fix(parser): ignore fenced code blocks when parsing delta specs

[javigomez]

Summary

Markdown structure written inside fenced code blocks was being parsed as
real content by the delta-spec parser (src/core/parsers/requirement-blocks.ts).
For example, documenting the delta format inside a scenario:

## ADDED Requirements

### Requirement: Documentation Generator
The system SHALL render a delta example in its output.

#### Scenario: Renders an example
**Given** a template
**When** documentation is generated
**Then** the following snippet is produced:

```markdown
### Requirement: Example only
#### Scenario: Example scenario
```

…caused ### Requirement: Example only to be parsed as a second, phantom
requirement. Because that phantom block has no body text, openspec validate
reported a spurious ADDED "Example only" is missing requirement text error,
and the same mis-parse feeds archive (specs-apply.ts), risking incorrect
spec output.

Root cause

Fenced-code detection already existed in MarkdownParser and spec-structure.ts
(and ChangeParser uses it), but it was duplicated in each place and
missing entirely from requirement-blocks.ts, which powers both validate
and archive.

Changes

• Add a single shared buildCodeFenceMask helper (src/core/parsers/code-fence.ts).
• Make parseDeltaSpec and extractRequirementsSection honor fenced code when
  detecting section/requirement/removed/renamed headers.
• Make the validator's countScenarios / extractRequirementText fence-aware too.
• Refactor MarkdownParser and spec-structure.ts to reuse the shared helper
  instead of their own copies (no behavior change there; removes the drift that
  caused this bug). Public/protected APIs are unchanged.

No new features, no config or output format changes.

Test plan

• New regression tests in test/core/parsers/requirement-blocks.test.ts
  (fenced requirement headers, REMOVED bullets and RENAMED pairs are ignored;
  main-spec requirements section too).
• New validator integration test in test/core/validation.test.ts
  (fenced example no longer triggers a spurious error).
• node build.js (tsc) passes.
• eslint src/ passes.
• Full suite green: 1659 passed (was 1655 + 4 new).
• Added a patch changeset.

Made with Cursor

Summary by CodeRabbit

• Bug Fixes
  • Improved Markdown parsing to correctly ignore requirement and delta structure inside fenced code blocks.
  • Prevented example content in fenced blocks from creating false requirements, validation errors, or incorrect archive results.
  • Standardized fenced-code handling across specification and change parsing.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 44ebd699-67f3-4e2d-a576-912bc79f3d05

📥 Commits

Reviewing files that changed from the base of the PR and between b6c7a86 and c248f66.

📒 Files selected for processing (1)

• src/core/parsers/requirement-blocks.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• src/core/parsers/requirement-blocks.ts

---

📝 Walkthrough

Walkthrough

Fence-aware parsing is centralized in a shared mask utility. Requirement, delta-spec, and fenced-content parsing now ignore Markdown structures inside code fences, with regression tests covering requirements, directives, renamed or removed entries, and validation scenarios.

Changes

Fence-aware Delta-spec Parsing

Layer / File(s)	Summary
Shared fence-detection utility src/core/parsers/code-fence.ts	Adds buildCodeFenceMask for tracking backtick and tilde fences and marking fenced lines.
Parser integration src/core/parsers/spec-structure.ts	Reuses the shared mask when stripping fenced blocks while preserving line positions.
Requirement and delta parsing src/core/parsers/requirement-blocks.ts	Makes section, requirement, removed, and renamed entry detection ignore fenced content.
Regression coverage and changeset test/core/parsers/requirement-blocks.test.ts, test/core/validation.test.ts, .changeset/fence-aware-delta-parsing.md	Adds fenced-content parsing and validation tests and documents the patch release.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Possibly related issues

• ### Requirement: inside a fenced code block is parsed as a real requirement #1348 — Directly concerns making requirement-section extraction ignore requirement headers inside fenced code blocks.

Possibly related PRs

• Fission-AI/OpenSpec#966 — Overlaps in making spec-structure parsing fence-aware.
• Fission-AI/OpenSpec#1031 — Overlaps in modifying requirement-header detection.
• Fission-AI/OpenSpec#1281 — Overlaps in ignoring fenced requirement and scenario markers.

Suggested reviewers: tabishb

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: making delta-spec parsing ignore fenced code blocks.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

test/core/validation.test.ts (1)

651-686: ⚡ Quick win

Add a fenced-only scenario regression here.

This test still passes if countScenarios() starts counting #### Scenario: lines inside fences again, because Line 663 already provides a real scenario. Add a sibling case with no unfenced scenario and assert the validator still reports the missing-scenario error.

Example follow-up case

+    it('does not count scenario headers inside fenced code blocks toward the required scenario count', async () => {
+      const changeDir = path.join(testDir, 'test-change-fenced-scenario-only');
+      const specsDir = path.join(changeDir, 'specs', 'test-spec');
+      await fs.mkdir(specsDir, { recursive: true });
+
+      const deltaSpec = `# Test Spec
+
+## ADDED Requirements
+
+### Requirement: Documentation Generator
+The system SHALL render a delta example in its output.
+
+\`\`\`markdown
+#### Scenario: Example scenario
+\`\`\`
+`;
+
+      const specPath = path.join(specsDir, 'spec.md');
+      await fs.writeFile(specPath, deltaSpec);
+
+      const report = await new Validator(true).validateChangeDeltaSpecs(changeDir);
+
+      expect(report.valid).toBe(false);
+      expect(
+        report.issues.some((i) => i.message.includes('must include at least one scenario'))
+      ).toBe(true);
+    });

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@test/core/validation.test.ts` around lines 651 - 686, The test currently only
covers a case where a fenced code block contains a faux requirement+scenario but
an unfenced scenario exists so a regression in countScenarios() could be missed;
add a sibling test case in test/core/validation.test.ts that writes a spec where
the only "#### Scenario:" is inside a fenced block (no unfenced scenario), call
Validator.validateChangeDeltaSpecs on that changeDir, and assert the validator
reports the missing-scenario error (e.g., report.valid is false,
report.summary.errors > 0 and report.issues includes a message about missing
scenario). This ensures countScenarios() and the Validator class logic correctly
ignore fenced code blocks and surface the missing-scenario error when
appropriate.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@test/core/validation.test.ts`:
- Around line 651-686: The test currently only covers a case where a fenced code
block contains a faux requirement+scenario but an unfenced scenario exists so a
regression in countScenarios() could be missed; add a sibling test case in
test/core/validation.test.ts that writes a spec where the only "#### Scenario:"
is inside a fenced block (no unfenced scenario), call
Validator.validateChangeDeltaSpecs on that changeDir, and assert the validator
reports the missing-scenario error (e.g., report.valid is false,
report.summary.errors > 0 and report.issues includes a message about missing
scenario). This ensures countScenarios() and the Validator class logic correctly
ignore fenced code blocks and surface the missing-scenario error when
appropriate.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 725c321b-78bf-4210-a01d-d12e1da4475b

📥 Commits

Reviewing files that changed from the base of the PR and between 055957f and 5d97411.

📒 Files selected for processing (8)

• .changeset/fence-aware-delta-parsing.md
• src/core/parsers/code-fence.ts
• src/core/parsers/markdown-parser.ts
• src/core/parsers/requirement-blocks.ts
• src/core/parsers/spec-structure.ts
• src/core/validation/validator.ts
• test/core/parsers/requirement-blocks.test.ts
• test/core/validation.test.ts

[alfred-openspec]

Reviewed the parser/validator paths and this looks good to merge.

The shared buildCodeFenceMask keeps the existing MarkdownParser/spec-structure behavior while wiring the missing fence awareness into requirement-blocks.ts, which is the important archive/validate path. I also checked the CodeRabbit scenario-count nit: the current implementation does ignore fenced-only #### Scenario: lines and still reports the missing-scenario error, so I do not think that extra test should block merge. It would still be a nice regression to add if you want the belt-and-suspenders coverage.

Local verification:

• pnpm exec vitest run test/core/parsers/requirement-blocks.test.ts test/core/validation.test.ts
• pnpm exec vitest run test/specs/source-specs-normalization.test.ts test/core/parsers/markdown-parser.test.ts test/core/parsers/change-parser.test.ts
• fenced-only scenario throwaway fixture against built validator
• pnpm run build
• pnpm run lint
• pnpm test → 1659 passed

[javigomez]

Addressed the scenario-count nit from the review.

Pushed b6c7a86, which adds a focused regression test in test/core/validation.test.ts. The existing fenced-example test always includes a real (unfenced) #### Scenario:, so it would still pass even if countScenarios() regressed and began counting scenario headers inside fenced code blocks. The new test covers the case where the only #### Scenario: lives inside a fenced block and asserts the validator still reports the must include at least one scenario error — locking in the fence-aware behavior @alfred-openspec confirmed during review.

Test-only change: no source/behavior changes, and no new changeset (the existing patch changeset already covers the fix).

Verification:

• pnpm test → 1660 passed (1659 + 1 new)
• pnpm exec tsc --noEmit → clean
• pnpm lint → clean

[alfred-openspec]

Re-reviewed after the latest commits and this is still ready to merge.

The new fenced-only scenario regression covers the previous CodeRabbit nit directly: fenced #### Scenario: examples no longer count toward the required scenario count, so real missing-scenario errors still fire. I re-ran the focused parser/validator tests and build locally on head 0816db3, all green.

[javigomez]

Hi, is there anything I can do on my side to move forward with this PR?

[alfred-openspec]

Thanks for checking in. Nothing needed from your side right now, this is already approved and mergeable. I’m leaving the actual merge to Tabish, but the PR itself looks ready.